CUDA 11.1 released with support for RTX30 series cards. I'm guessing that the ACEMD app will need to be recompiled with the latest toolkit to support the Ampere cards (if people can actually get one).

Full story here

---

BOINC blog

The cards seem to work on the existing apps just fine. At least for Milkyway, Einstein and Primegrid. No recompilation was needed.

Whether the existing apps are getting all they can out of the hardware is a different question.

CUDA 11.1 also brings a new PTX compiler static library, version 7.1 of the Parallel Thread Execution (PTX) ISA, support for Fedora 32 and Debian 10.3, new unified programming models, hardware-accelerated sparse texture support, multi-threaded launch to different CUDA streams, improvements to CUDA Graphs, and various other enhancements. GCC 10.0 and Clang 10.0 are also now supported as host compilers.

That does suggest that there is more performance to be squeezed out of the hardware once the software is fine-tuned.

I'd like to see a CUDA 11.1 Linux app here at GPUGRID :)

---

That does suggest that there is more performance to be squeezed out of the hardware once the software is fine-tuned.

I'd like to see a CUDA 11.1 Linux app here at GPUGRID :)

When everything seems to have settled, always a new challenge arises...

Toni mentioned that they also need the OpenMM libraries to be updated before they can update the app.

---

BOINC blog

When everything seems to have settled, always a new challenge arises...

As Heraclitus said "Change is the only constant in life."

---

BOINC blog

Toni mentioned that they also need the OpenMM libraries to be updated before they can update the app.

Can you post Toni's post number, please.

Toni mentioned that they also need the OpenMM libraries to be updated before they can update the app.

Can you post Toni's post number, please.

I sent him a PM telling him about CUDA 11.1. He mentioned it in his reply.

A quick google search led me to OpenMM.org and they have a bug raised for Ampere performance here which seems to indicate they don't think it needs a new version, so maybe its just a matter of recompiling with CUDA 11.1.

---

BOINC blog

Thanks for the reply. Following the thread through the OpenMM link was illuminating.

Will watch for Toni's announcement of a new app eventually.

Thanks for the reply. Following the thread through the OpenMM link was illuminating.

I'll second that, Keith. Thanks for that MarkJ!

with the 3060ti release today, that's 4 models of Ampere cards that will be floating around in the wild.

GPUGRID, Einstein, and Milkyway are the only projects that have GPU app and aren't some number theory thought experiment. out of these GPUGRID is the only one providing a CUDA app. Currenty Einstein is the only worthwhile project to contribute to on a 30-series card since MW requires DP and DP is quite poor on the 30-series cards. The Einstein app is 'ok' but certainly favors AMD for productivity/watt since they lack a CUDA optimized app.

any update on the progress of a CUDA 11.1 app here? We need a place for the 30-series cards to be more productive. My 3070 is running at Einstein, but I want to get it here to stretch its legs.

---

with the 3060ti release today, that's 4 models of Ampere cards that will be floating around in the wild.

GPUGRID, Einstein, and Milkyway are the only projects that have GPU app and aren't some number theory thought experiment. out of these GPUGRID is the only one providing a CUDA app. Currenty Einstein is the only worthwhile project to contribute to on a 30-series card since MW requires DP and DP is quite poor on the 30-series cards. The Einstein app is 'ok' but certainly favors AMD for productivity/watt since they lack a CUDA optimized app.

any update on the progress of a CUDA 11.1 app here? We need a place for the 30-series cards to be more productive. My 3070 is running at Einstein, but I want to get it here to stretch its legs.

Just to add to this - I have a 3070 and a 3080 that would be crunching GPUGrid full time at least until the end of the year but probably much longer if there was an app available. I would have been running GPUGrid on them since I got them if it functioned. I think enough cards are out there now that a lot of users would have them and be running them on GPUGrid.